SERMAS launches dxGPT, a collaboration with Microsoft to incorporate AI in diagnosis
First of all, a warning. “dxGPT is a research experiment,” those responsible for the project warned when they opened their official website. This system uses OpenAI’s GPT-4 linguistic model as the basis of a tool with one goal: to help doctors give a diagnosis to patients.
Madrid, pioneer. The Madrid Health Service (SERMAS) has become “the first of its kind in Spain and in the whole world” to launch a project of this kind, they say to Microsoft. It was launched by the Redmond giant with the Community of Madrid – Department of Digitalization – and Fundación 29, and the goal is to improve the health care of patients and facilitate the work of medical staff.
A ChatGPT for doctors. This chatbot is in the form of a web application that in its pilot phase will be available in Primary Care centers in the Community of Madrid at the end of September. It is based on GPT-4, and Microsoft provides the infrastructure through the Azure OpenAI Service cloud platform.
Rare diseases. The proposal is essentially aimed at helping the use of Generative AI technologies “for the clinical diagnosis of rare diseases.” On average, these diseases take “about five years to have a correct diagnosis” and they have to visit an average of up to seven specialists to reach that diagnosis. Despite all this, more than half of these patients did not even achieve a diagnosis or ended up with an error.
An assistant to specialists. With DxGPT the goal is to offer accurate assistance to medical professionals allowing them to reduce the diagnosis time to minutes. This tool, emphasized Microsoft, “does not replace the doctor, but improves his capabilities and helps him do his job in less time.”
You say it with symptoms, DxGPT gives you diagnostic options. The operation is the same as ChatGPT: in a text box the patient or specialist can enter a description of the symptoms, their age and state of health, and from there in a few seconds the chatbot will offer a series of possible diagnoses. To improve the results, it is possible to increase the symptoms or provide information on what steps should be taken next.
The robots that whispered to the doctors. This project is a pioneer in Madrid, but the use of generative AI in the field of medicine is not new. Google introduced Med-PaLM in 2022 and its second version in March 2023. In both cases the basis is the same: a large language model (LLM) specifically trained on medical literature to provide answer addressed in this section. According to those responsible, Med-Palm 2 can achieve an accuracy of 86.5% in specific tests to evaluate the accuracy of medical diagnosis.
A dangerous project. It’s one thing to ask ChatGPT to write a paper about Don Quixote and another to trust it with our sanity. These types of machines have been shown to make mistakes and hallucinate, but their answers are usually given with such conviction that one can assume they are correct without much doubt.
Management, essential. Precisely in the field of medicine and diagnosis it is very dangerous, but Microsoft assures that the application “has all the standards to improve security and ethics.” The website insists on this and indicates that “it is important to consult a qualified health professional to ensure an accurate and appropriate diagnosis.” The creators of the web application indicate that the use of the tool “is at the user’s own risk and Fundación 29 is not responsible for the consequences or damages that may arise from its use.”